1. Algorithm modification: Instead of having same filter threshold for a whole frame, now we allow the thresholds to be adjusted for each macroblock. In current implementation, to avoid excessive blur on background as reported in issue480(http://code.google.com/p/webm/issues/detail?id=480), we reduce the thresholds for skipped macroblocks. 2. SSE2 optimization: As started in issue479(http://code.google.com/p/webm/issues/detail?id=479), the filter calculation was adjusted for better performance. The c code was also modified accordingly. This made the deblock filter 2x faster, and the decoder was 1.2x faster overall. Next, the demacroblock filter will be modified similarly.
Showing the changes to test (1 of 8 files). Show all
Unified · Split
| @@ -19,9 +19,9 @@ typedef void (*post_proc_func_t)(unsigned char *src_ptr, · 18 unmodified lines | ||
| 19 | 19 | unsigned char *dst_ptr, |
| 20 | 20 | int src_pixels_per_line, |
| 21 | 21 | int dst_pixels_per_line, |
| 22 | − int rows, | |
| 23 | 22 | int cols, |
| 24 | − int flimit); | |
| 23 | + unsigned char *flimit, | |
| 24 | + int size); | |
| 25 | 25 | |
| 26 | 26 | namespace { |
| 27 | 27 | |
| @@ -29,7 +29,7 @@ class Vp8PostProcessingFilterTest · 1 unmodified line | ||
| 29 | 29 | : public ::testing::TestWithParam<post_proc_func_t> {}; |
| 30 | 30 | |
| 31 | 31 | // Test routine for the VP8 post-processing function |
| 32 | −// vp8_post_proc_down_and_across_c. | |
| 32 | +// vp8_post_proc_down_and_across_mb_row_c. | |
| 33 | 33 | |
| 34 | 34 | TEST_P(Vp8PostProcessingFilterTest, FilterOutputCheck) { |
| 35 | 35 | // Size of the underlying data block that will be filtered. |
| @@ -56,6 +56,8 @@ TEST_P(Vp8PostProcessingFilterTest, FilterOutputCheck) { · 20 unmodified lines | ||
| 56 | 56 | // Pointers to top-left pixel of block in the input and output images. |
| 57 | 57 | uint8_t *const src_image_ptr = src_image + (input_stride << 1); |
| 58 | 58 | uint8_t *const dst_image_ptr = dst_image + 8; |
| 59 | + uint8_t *const flimits = reinterpret_cast<uint8_t *>(vpx_memalign(16, block_width)); | |
| 60 | + (void)vpx_memset(flimits, 255, block_width); | |
| 59 | 61 | |
| 60 | 62 | // Initialize pixels in the input: |
| 61 | 63 | // block pixels to value 1, |
| @@ -73,14 +75,13 @@ TEST_P(Vp8PostProcessingFilterTest, FilterOutputCheck) { · 11 unmodified lines | ||
| 73 | 75 | (void)vpx_memset(dst_image, 99, output_size); |
| 74 | 76 | |
| 75 | 77 | GetParam()(src_image_ptr, dst_image_ptr, input_stride, |
| 76 | − output_stride, block_height, block_width, | |
| 77 | − 255); | |
| 78 | + output_stride, block_width, flimits, 16); | |
| 78 | 79 | |
| 79 | 80 | static const uint8_t expected_data[block_height] = { |
| 80 | − 3, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 3 | |
| 81 | + 4, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 3, 4 | |
| 81 | 82 | }; |
| 82 | 83 | |
| 83 | − pixel_ptr = dst_image; | |
| 84 | + pixel_ptr = dst_image_ptr; | |
| 84 | 85 | for (int i = 0; i < block_height; ++i) { |
| 85 | 86 | for (int j = 0; j < block_width; ++j) { |
| 86 | 87 | EXPECT_EQ(expected_data[i], pixel_ptr[j]) |
| @@ -91,19 +92,15 @@ TEST_P(Vp8PostProcessingFilterTest, FilterOutputCheck) { · 4 unmodified lines | ||
| 91 | 92 | |
| 92 | 93 | vpx_free(src_image); |
| 93 | 94 | vpx_free(dst_image); |
| 95 | + vpx_free(flimits); | |
| 94 | 96 | }; |
| 95 | 97 | |
| 96 | 98 | INSTANTIATE_TEST_CASE_P(C, Vp8PostProcessingFilterTest, |
| 97 | − ::testing::Values(vp8_post_proc_down_and_across_c)); | |
| 98 | − | |
| 99 | −#if HAVE_MMX | |
| 100 | −INSTANTIATE_TEST_CASE_P(MMX, Vp8PostProcessingFilterTest, | |
| 101 | − ::testing::Values(vp8_post_proc_down_and_across_mmx)); | |
| 102 | −#endif | |
| 99 | + ::testing::Values(vp8_post_proc_down_and_across_mb_row_c)); | |
| 103 | 100 | |
| 104 | 101 | #if HAVE_SSE2 |
| 105 | 102 | INSTANTIATE_TEST_CASE_P(SSE2, Vp8PostProcessingFilterTest, |
| 106 | − ::testing::Values(vp8_post_proc_down_and_across_xmm)); | |
| 103 | + ::testing::Values(vp8_post_proc_down_and_across_mb_row_sse2)); | |
| 107 | 104 | #endif |
| 108 | 105 | |
| 109 | 106 | } // namespace |